It is important to estimate the banana ripeness and prolong the banana shelf life.Firstly, in recent years, the UV-C (Ultraviolet-C light) as safe, green method to replace the traditional thermal sterilization, becomes popular in circulation of fruits and vegetables.UV-C treatment could improve the disease resistance of the fresh agricultural products, delay senescence, inhibit pathogenic microorganisms breeding.So it becomes the research boiling point in the harvest and storage of fresh agricultural products.in this paper the trial bench of treatment banana was established in order to generate different UV-C (Ultraviolet-C light) dosages.Bothsides of the banana fruits were illuminated with UV-C each at thenominal illumination duration to obtain irradiation uniformity as followed in Fig.1.Due to different light distancesthe dosages were 55, 43,37, 29, 24, 21, 18,15μW/cm2,respectively.Theillumination dosages were determined with a photo-radiometer(ZQJ-254, CHINA).The visual qualities were analyzed through the image,the chromatic aberration test, the firmness test and the black spot of banana skin test, combination with no UV-C treatment analysis.The results showedthe high dosages UV-C damage to the banana skin, while the low dosages UV-C treatment had better efficiency and can postpone the change of the relative brightness.Safedosage UV-C treatments can delay the change of the relative color aberration, A, B, promote the firmness of the banana and retard the generation of black spot.The safe UV-C treatment doses were21 and 24μW/cm2 and it can prolong the shelf life of the banana.Secondly, Banana fruit surface and maturity stages were studied by using hyperspectral imaging technique in the visible and near infrared (370-1000 nm) regions.Artificial Neural Networks (ANNs) are known to be suitablefor finding unknown correlation between a given input datasetand its target set.In the present work a Feed Forward Back-propagation (FFBP) ANN has been devised in order torefine the prediction of hyperspectral data.As widely known ANNs requiretraining in order to "learn" how to correlate inputs with outputs.Accordingly proper datasets to accomplish the training phase have to be selected.Particularly, our selected colorratios from these spectral images were used for classifying the whole banana into immature, ripe, half-ripe,andoverripe stages.By using BP neural network, models were established based on the wavelengths to predict the quality attributes.The random 30 samples were drawn as training samples.The other 30 samples were as testing samples.For reduces errors, the training samples and the test samples were exchanged for 20 times.The mean discrimination rate is 98.17%.The target outputs and real outputs were as table 1.This paper experimentally howed ahyperspectral-imaging with a color feature extraction algorithm and artificial neural network were proposed for classifying maturity levels of the banana.The results improved the visual quality for the classification of the banana.